PHENOTYPIC CORRELATION AND PATH ANALYSIS IN CULTIVARS AND STRAINS OF UPLAND RICE FOR DROUGHT TOLERANCE CORRELAÇÃO FENOTÍPICA E ANÁLISE DE TRILHA EM CULTIVARES E LINHAGENS DE ARROZ DE SEQUEIRO PARA TOLERÂNCIA À SECA

The purpose of this study was to estimate the phenotypic correlations between 14 traits obtained in a thematic core collection of upland rice for drought tolerance and partition them into direct and indirect effects by path analysis. Two experiments were carried out (with and without water stress). One hundred samples were evaluated in a triple 10x10 lattice design. The plot was formed by four rows, 3.0 metres long, spaced at 0.35 m. The plot useful area was constituted by two central rows of 2.0 m in length, totalling 1.4 m, where data from 14 traits were collected, five from the root system and nine from the aerial part of the plant. Of the evaluated traits, spikelet sterility was the main grain yield determinant, presenting relevant negative correlations of -0.77 and -0.59 in environments with and without drought stress, respectively. The partitioning of spikelet sterility correlations presented negative direct effects on grain yield in environments with (-0.60) and without (-0.62) water stress, corroborating the negative correlations between these traits. The obtained data confirmed that spikelet sterility is an important variable for the selection of rice strain submitted to water deficit. Partial correlation coefficients indicated that only 70.33% in the environment with stress and 50.30% in the environment without stress of grain yield variation were phenotypically explained by variables considered in path analysis, thereby showing the complexity of the selection for drought-tolerant rice.


INTRODUCTION
Rice is one of the main cereals cultivated in the world and its evolution process led it to adapt to the most varied soil and climate conditions. Planting conditions vary from flooded environments, such as in the case of flooded rice systems, where, in general, varieties of the indica subtype are used, to dry environments (upland rice). The water necessary for plant growth and development in the upland system depends on the rainfall regime, and this system is therefore vulnerable to drought stress. In this latter cultivation system, varieties of the japonica subtype are used (ABADIE et al., 2005).
The stress caused by water deficit has long caused yield reduction in several crops, and this has recently been aggravated due to climate changes. Under this type of stress, the plant generally shows an increase in diffusive resistance to water vapour, thanks to stomatal closure, a reduction in transpiration and carbon dioxide supply for the photosynthetic process, reduced cellular growth and increased photorespiration (SHINOZAKI; YAMAGUCHI-SHINOZAKI, 2007), thereby reducing the photosynthesis rate and, consequently, having undesirable effects on vigour and plant height, pollen grain fertility and yield (BOTA et al., 2004). For Nguyen et al. (1997), drought resistance physiologic mechanisms in rice cultivation are related to moderate water use by the plant, leaf area reduction, leaf water loss control and root ability in exploring deeper soil layers. An improvement in yield stability in environments with water deficit is essential for this cultivation, and can be carried out by identifying the traits that may contribute to drought resistance (BABU et al., 2003).
The development of water stress-tolerant upland rice cultivars by breeding is recognized as the most efficient strategy to relieve food insecurity caused by water shortage (HUANG et al., 2007). However, little success has been achieved so far in developing water stress-tolerant cultivars due to this characteristic having a quantitative heritage and strong environmental influence (SHINOZAKI; YAMAGUCHI-SHINOZAKI, 2007).
Traits controlled by several genes, with low heritability and great interaction with the environment, can be difficult to measure, and, therefore, have low efficiency upon selection. One of the alternatives to increase gains by selection in a low-heritage characteristic would be to select another characteristic of high heritage, easy mensuration and strong correlation (LAFITTE et al., 2003). The two causes of genetic correlation between two traits are pleiotropy and genic connection. The correlation caused by pleiotropy is when two characters are influenced by the same genes (KUREK et al., 2001), which are of great importance in breeding programmes (MARCHEZAN et al., 2005). However, this simple correlation may induce mistakes in the relation between two variables, as there may be an effect of a third variable acting on the expression of the main variable. To overcome this problem, WRIGHT (1921) proposed partitioning the correlations between characters into direct and indirect effects, using path analysis (CRUZ et al., 2004). Path analysis is a standardized partial regression coefficient that allows the correlation coefficient to be partitioned into direct and indirect effects and the action of specific components that produce a certain correlation between variables to be studied.
Understanding the relations between the traits that influence water stress resistance is of vital importance for genetic improvement programmes aimed at developing upland rice cultivars tolerance to this abiotic stress (ABREU et al., 2016). Thus, the purpose of the present study was to estimate the phenotypic correlations between 14 traits obtained in a thematic core collection of upland rice for drought resistance and to partition them into direct and indirect effects by path analysis.

MATERIAL AND METHODS
Two experiments (with and without water stress) were carried out between June and November 2007 in the Experimental Unit of the Federal University of Tocantins (UFT), at the University Campus of Gurupi, TO, Brazil, situated at a latitude of 11.7458 º south, a longitude of 49.0497 º west and an altitude of 280 m. In this region, a well-defined period without rainfall occurs, from May to September, which represents a favourable environment for studies on drought resistance. The rainfall from June to October was practically nil, beginning in the second half of October with low intensity (Figure 1). Maximum and minimum temperatures were approximately 35 °C and 15 °C, respectively, from June to August and 20 °C from September to November. The relative air moisture in this period varied from very low values of approximately 10% to the highest value of 90%.
The soil where the experiments were carried out is characterized as a dystrophic yellow latosol with a sandy texture. Starter and top dressing fertilization, as well as weed, pest and disease control, were carried out according to technical recommendations for upland rice farming.
Both experiments (with and without water stress) were equally irrigated up to 35 days after emergence (DAE), using an irrigation system formed by a self-propelled set and wheeled vehicle with side bars and regulating pressure valves at the end of spraying tips, to obtain a constant and homogeneous water sheet during application. After this period, for the stress treatment, approximately half of the water sheet was supplied, considering that the soil water tension used in the experiment for the environment without drought stress was 250 mb, as suggested by Stone et al. (1986). Irrigation shift was adjusted according to crop evapotranspiration, using tensiometers placed in strategic areas, with the porous capsule at a depth of 0.15 m in the soil. One hundred samples were evaluated, all of the japonica subtype, with 86 traditional rice varieties from the Thematic Core Collection of Upland Rice for Drought Tolerance (TERRA et al., 2015) and 14 cultivars and elite strains from the programme for genetic improvement of Embrapa The following data were collected: a) Leaf canopy temperature (LCT), obtained from two samples per plot in each measurement, totalling five readings per treatment. Evaluations were carried out using a laser thermometer (Raytek Raynger ST), placed at 10 cm of leaf canopy, always on the day prior to irrigation, at the hottest time of the day (between 12:30 h and 14:00 h), with a more elevated level of soil water tension, in both treatments (with and without stress); b) Flowering average (FLO), number of days from planting to 50% tillers with panicles; c) Plant height (H), obtained at harvesting from 10 competitive plants in the useful plot area; d) Number of tillers (TL); e) Percentage of tiller sterility (TS), obtained by counting the number of tillers and panicles in a linear metre, and the number and percentage of tillers without panicles verified; f) Number of filled grains per panicle (GPP); g) Percentage of spikelet sterility (PSS) per panicle; h) 100-grain weight (GW), obtained from a sample of 10 panicles by plot by counting the number of full and empty grains; i) The percentage of spikelet sterility is considered a primary component of production (ALVAREZ et al., 2012) and was obtained using the following formula: PSS = (EG x 100) / TS; where: EG = number of empty grains; TS = total number of spikelets per panicle; j) Grain yield in 1.4 m 2 (Y) of evaluated genotypes was obtained from the harvest of grains from two central rows from each plot, discarding 0.5 m from the end of each row, totalling a useful harvest area of 1.4 m 2 ; k) Root data, collected at early flowering in each plot. For this, soil and root samples were collected at depths of 20 to 60 cm, using a drill with a 7 cm internal diameter. These samples were removed at a distance of approximately 7 cm from one of the two central rows. The first 20 cm were discarded. The soil and root samples were washed in water and sieved in 1 mm sieves, separating the roots. Then the samples were placed in plastic bags, properly identified and stored in a freezer. Root measurement and qualification was carried out using WinRHIZO (Regent Instruments Canada INC, 2008) Pro v 2008a, 32-bit software. For that, root samples were thawed and placed in acrylic tray with a water line/sheet of 2 to 3 mm, and then scanned. Images were stored in TIFF format. The root variables estimated by the WinRHIZO software were: root length (RL), contact surface (CS), mean root diameter (MRD), volume (V) and number of root branches (RB).
Phenotypic correlation coefficients were estimated by variance and covariance components, according to the method suggested by Kempthorne (1973), Mather (1966), and Mode and Robinson (1959). The significance of the correlation coefficients was evaluated by t-test at 1% and 5% probability.
Phenotypic correlation coefficients were partitioned into direct and indirect effects using the method of path coefficients, showing correlations for grain yield (Y) in 1.4 m 2 as the main variable, and FLO, H, GPP, PSS, GW, TL and TS, LCT, RL, CS, MRD, V and RB as explanatory variables.
The analyses of data were carried out using the Genes version 2013 5.0.1 software program for statistical analyses (CRUZ, 2013). Table 1 shows estimates of phenotypic correlations between all evaluated variables for two farming conditions, with and without water stress. Significant differences were observed at 1% and 5% probability by t-test for several phenotypic correlation coefficients.

Phenotypic correlations
Negative correlations were obtained between the variables Y and PSS of -0.77 and -0.59 in environments with and without stress, respectively. This shows that spikelet sterility in panicles is the main determinant of grain yield. Similar results were obtained by LAFITTE et al. (2003) and KUMAR et al. (2008). These authors showed that due to higher PSS inheritance, it could provide higher selection gains.
In the condition of water stress, the variable Y was positively correlated with the variables H (0.31**), GPP (0.27**) and GW (0.45**). The same variable was negatively correlated with TL (-0.33**). Conversely, in the condition without stress, Y presented a positive and significant correlation only with GW (0.35**) and a negative and significant correlation with the variables FLO (-0.31**) and H (-0.21**). Y was positively correlated with H (0.31**) in the condition with stress and negatively (-0.21*) in the condition without stress.  The results show a tendency of higher plants to present higher drought resistance. This may be due to the fact that plants with this morphology present a greater root volume in deeper soil layers, and consequently can explore wetter areas. In the definition of a plant ideotype for upland rice conditions, this characteristic needs to be considered, prioritizing higher plants.
FLO presented, in conditions with and without stress, positive and significant correlations with the variable H (0.67** and 0.76**, respectively). These correlations might create some difficulty in obtaining more precocious varieties (average flowering around 70 days) and higher plants when using a direct planting system, where upland rice is the second option after soy. Lafitte et al. (2003) reported that the number of days till flowering is a trait that can be used to differentiate genotypes with different responses to drought stress, and that the difference between flowerings may vary from 12 days before to 7 days after, when compared with the condition without stress.
The H variable presented positive and significant correlations with GPP (0.57** and 0.32** with and without stress, respectively) in two cultivation conditions, showing that higher plants, in general, have larger panicles and more grains.
Leaf canopy temperature (LCT) was not correlated with any traits of the aerial part evaluated in both environments in the present study. This shows that this variable is of little use as one of the determining drought resistance parameters in the evaluated population.
As for root parameters (RL, CS, MRD and V), correlations with aerial part variables in the environment with stress were small, even those that presented 1% and 5% significance by t-test. Root data collection in the field or even in controlled conditions is a laborious process and in some cases it is not completely accurate, due to the methodologies used, soil separation process or root quantification. This may in part explain the decreased magnitudes of correlations with the aerial plant part. In the environment without stress, the most elevated observed correlations were between RL x H (0.5228**) and CS x H (0.5102 **), indicating that higher plants can present a greater length and root contact surface with the soil.
In general, the correlations between root variables were elevated and highly significant, which was expected, since the evaluated root traits are not independent. Tables 2 and 3 show the results for  phenotypic correlation coefficients for traits FLO,  H, GPP, PSS, GW, TL, TS, LCT, RL, CS, MRD, V and RB with grain yield in 1.4 m 2 .

Path analysis
According to Jagadish et al. (2007), spikelet sterility is a variable of great importance among the production primary components in rice farming and it is greatly affected by abiotic factors such as temperature and moisture. In this study, PSS presented negative direct effects on grain yield in environments with (-0.6051) and without (-0.62) stress, thereby corroborating the negative correlations between these traits (Table 2). PSS was the most important variable in the determination of grain yield in the evaluated genotype population. Jongdee et al. (2006) also emphasized that spikelet sterility is an important variable for the selection of rice strain submitted to water deficit.
In general, the direct and indirect effects of the traits FLO, H, GPP, GW, TL and TS were small, showing little influence of these variables on grain yield and corroborating the low estimate values obtained for phenotypic correlations.
The non-significant phenotypic correlations between leaf canopy temperature (LCT) and the evaluated traits (Table 1) are explained by the direct and indirect effects of low significance (Table 3), showing little importance of this variable in the selection of more drought-tolerant genotypes in the evaluated population.
In the environment with stress, root length and volume presented negative direct effects of -0.6741 and -0.7851, respectively, for grain yield. However, these effects were neutralized by elevated positive indirect effects of 1.0706 and 1.1294, respectively, for root surface (Table 3).  With regard to the root traits, the root contact surface (CS) in the environment with stress (Table 3) was the variable with the highest positive influence on grain yield, indicating that accesses with greater root contact surface with the soil would have a higher water absorption capacity. However, this direct effect (1.1943) was counterbalanced in part by the negative indirect effect for root volume (-0.7425).
The coefficients of partial determinations indicated that only 70.33% in the environment with stress and 50.30% in the environment without stress of the grain yield variation were explained, phenotypically, by variables considered in the path analysis. This shows the complexity of the selection for drought-tolerant rice, corroborated by the residual variation effect of 54.47% and 70.50% in the environments with and without water stress, respectively. The Experimental Field of the Federal University of Tocantins (Gurupi, TO, Brazil) is located in a region where a dry period occurs from May to September, when the experimental tests were carried out, allowing tests to be performed with and without water stress using irrigation. However, the elevated temperatures and low relative air moisture in determined phases of the rice cycle, mainly at flowering, induced sample damage due to the emission of white panicles.

CONCLUSIONS
Spikelet sterility was the trait with the highest correlations with productivity in both the conditions with and without stress, followed by grain weight.